opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/ShardCommitCoordinator.java

   1 /*
   2  * Copyright (c) 2014 Brocade Communications Systems, Inc. and others.  All rights reserved.
   3  *
   4  * This program and the accompanying materials are made available under the
   5  * terms of the Eclipse Public License v1.0 which accompanies this distribution,
   6  * and is available at http://www.eclipse.org/legal/epl-v10.html
   7  */
   8 package org.opendaylight.controller.cluster.datastore;
   9
  10 import static java.util.Objects.requireNonNull;
  11
  12 import akka.actor.ActorRef;
  13 import akka.actor.Status.Failure;
  14 import akka.serialization.Serialization;
  15 import com.google.common.annotations.VisibleForTesting;
  16 import com.google.common.primitives.UnsignedLong;
  17 import com.google.common.util.concurrent.FutureCallback;
  18 import edu.umd.cs.findbugs.annotations.SuppressFBWarnings;
  19 import java.util.ArrayDeque;
  20 import java.util.ArrayList;
  21 import java.util.Collection;
  22 import java.util.Collections;
  23 import java.util.Deque;
  24 import java.util.HashMap;
  25 import java.util.LinkedList;
  26 import java.util.Map;
  27 import org.eclipse.jdt.annotation.NonNull;
  28 import org.opendaylight.controller.cluster.access.concepts.TransactionIdentifier;
  29 import org.opendaylight.controller.cluster.datastore.messages.AbortTransactionReply;
  30 import org.opendaylight.controller.cluster.datastore.messages.BatchedModifications;
  31 import org.opendaylight.controller.cluster.datastore.messages.BatchedModificationsReply;
  32 import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction;
  33 import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransactionReply;
  34 import org.opendaylight.controller.cluster.datastore.messages.CommitTransaction;
  35 import org.opendaylight.controller.cluster.datastore.messages.CommitTransactionReply;
  36 import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction;
  37 import org.opendaylight.controller.cluster.datastore.messages.ReadyLocalTransaction;
  38 import org.opendaylight.controller.cluster.datastore.messages.ReadyTransactionReply;
  39 import org.opendaylight.controller.cluster.datastore.messages.VersionedExternalizableMessage;
  40 import org.opendaylight.controller.cluster.datastore.utils.AbstractBatchedModificationsCursor;
  41 import org.opendaylight.yangtools.concepts.Identifier;
  42 import org.opendaylight.yangtools.yang.common.Empty;
  43 import org.opendaylight.yangtools.yang.data.tree.api.DataTreeCandidate;
  44 import org.slf4j.Logger;
  45
  46 /**
  47  * Coordinates commits for a shard ensuring only one concurrent 3-phase commit.
  48  *
  49  * @author Thomas Pantelis
  50  */
  51 final class ShardCommitCoordinator {
  52
  53     // Interface hook for unit tests to replace or decorate the ShardDataTreeCohorts.
  54     @VisibleForTesting
  55     public interface CohortDecorator {
  56         ShardDataTreeCohort decorate(Identifier transactionID, ShardDataTreeCohort actual);
  57     }
  58
  59     private final Map<Identifier, CohortEntry> cohortCache = new HashMap<>();
  60
  61     private final ShardDataTree dataTree;
  62
  63     private final Logger log;
  64
  65     private final String name;
  66
  67     // This is a hook for unit tests to replace or decorate the ShardDataTreeCohorts.
  68     @VisibleForTesting
  69     private CohortDecorator cohortDecorator;
  70
  71     private ReadyTransactionReply readyTransactionReply;
  72
  73     ShardCommitCoordinator(final ShardDataTree dataTree, final Logger log, final String name) {
  74         this.log = log;
  75         this.name = name;
  76         this.dataTree = requireNonNull(dataTree);
  77     }
  78
  79     int getCohortCacheSize() {
  80         return cohortCache.size();
  81     }
  82
  83     private String persistenceId() {
  84         return dataTree.logContext();
  85     }
  86
  87     private ReadyTransactionReply readyTransactionReply(final ActorRef cohort) {
  88         if (readyTransactionReply == null) {
  89             readyTransactionReply = new ReadyTransactionReply(Serialization.serializedActorPath(cohort));
  90         }
  91
  92         return readyTransactionReply;
  93     }
  94
  95     /**
  96      * This method is called to ready a transaction that was prepared by ShardTransaction actor. It caches
  97      * the prepared cohort entry for the given transactions ID in preparation for the subsequent 3-phase commit.
  98      *
  99      * @param ready the ForwardedReadyTransaction message to process
 100      * @param sender the sender of the message
 101      * @param shard the transaction's shard actor
 102      */
 103     void handleForwardedReadyTransaction(final ForwardedReadyTransaction ready, final ActorRef sender,
 104             final Shard shard) {
 105         log.debug("{}: Readying transaction {}, client version {}", name,
 106                 ready.getTransactionId(), ready.getTxnClientVersion());
 107
 108         final ShardDataTreeCohort cohort = ready.getTransaction().ready(ready.getParticipatingShardNames());
 109         final CohortEntry cohortEntry = CohortEntry.createReady(cohort, ready.getTxnClientVersion());
 110         cohortCache.put(cohortEntry.getTransactionId(), cohortEntry);
 111
 112         if (ready.isDoImmediateCommit()) {
 113             cohortEntry.setDoImmediateCommit(true);
 114             cohortEntry.setReplySender(sender);
 115             cohortEntry.setShard(shard);
 116             handleCanCommit(cohortEntry);
 117         } else {
 118             // The caller does not want immediate commit - the 3-phase commit will be coordinated by the
 119             // front-end so send back a ReadyTransactionReply with our actor path.
 120             sender.tell(readyTransactionReply(shard.self()), shard.self());
 121         }
 122     }
 123
 124     /**
 125      * This method handles a BatchedModifications message for a transaction being prepared directly on the
 126      * Shard actor instead of via a ShardTransaction actor. If there's no currently cached
 127      * DOMStoreWriteTransaction, one is created. The batched modifications are applied to the write Tx. If
 128      * the BatchedModifications is ready to commit then a DOMStoreThreePhaseCommitCohort is created.
 129      *
 130      * @param batched the BatchedModifications message to process
 131      * @param sender the sender of the message
 132      */
 133     @SuppressFBWarnings(value = "THROWS_METHOD_THROWS_RUNTIMEEXCEPTION", justification = "Replay of captured failure")
 134     void handleBatchedModifications(final BatchedModifications batched, final ActorRef sender, final Shard shard) {
 135         CohortEntry cohortEntry = cohortCache.get(batched.getTransactionId());
 136         if (cohortEntry == null || cohortEntry.isSealed()) {
 137             cohortEntry = CohortEntry.createOpen(dataTree.newReadWriteTransaction(batched.getTransactionId()),
 138                 batched.getVersion());
 139             cohortCache.put(cohortEntry.getTransactionId(), cohortEntry);
 140         }
 141
 142         if (log.isDebugEnabled()) {
 143             log.debug("{}: Applying {} batched modifications for Tx {}", name,
 144                     batched.getModifications().size(), batched.getTransactionId());
 145         }
 146
 147         cohortEntry.applyModifications(batched.getModifications());
 148
 149         if (batched.isReady()) {
 150             if (cohortEntry.getLastBatchedModificationsException() != null) {
 151                 cohortCache.remove(cohortEntry.getTransactionId());
 152                 throw cohortEntry.getLastBatchedModificationsException();
 153             }
 154
 155             if (cohortEntry.getTotalBatchedModificationsReceived() != batched.getTotalMessagesSent()) {
 156                 cohortCache.remove(cohortEntry.getTransactionId());
 157                 throw new IllegalStateException(String.format(
 158                         "The total number of batched messages received %d does not match the number sent %d",
 159                         cohortEntry.getTotalBatchedModificationsReceived(), batched.getTotalMessagesSent()));
 160             }
 161
 162             if (log.isDebugEnabled()) {
 163                 log.debug("{}: Readying Tx {} of {} operations, client version {}", name,
 164                         batched.getTransactionId(), cohortEntry.getTotalOperationsProcessed(), batched.getVersion());
 165             }
 166
 167             cohortEntry.setDoImmediateCommit(batched.isDoCommitOnReady());
 168             cohortEntry.ready(batched.getParticipatingShardNames(), cohortDecorator);
 169
 170             if (batched.isDoCommitOnReady()) {
 171                 cohortEntry.setReplySender(sender);
 172                 cohortEntry.setShard(shard);
 173                 handleCanCommit(cohortEntry);
 174             } else {
 175                 sender.tell(readyTransactionReply(shard.self()), shard.self());
 176             }
 177         } else {
 178             sender.tell(new BatchedModificationsReply(batched.getModifications().size()), shard.self());
 179         }
 180     }
 181
 182     /**
 183      * This method handles {@link ReadyLocalTransaction} message. All transaction modifications have
 184      * been prepared beforehand by the sender and we just need to drive them through into the
 185      * dataTree.
 186      *
 187      * @param message the ReadyLocalTransaction message to process
 188      * @param sender the sender of the message
 189      * @param shard the transaction's shard actor
 190      */
 191     void handleReadyLocalTransaction(final ReadyLocalTransaction message, final ActorRef sender, final Shard shard) {
 192         final TransactionIdentifier txId = message.getTransactionId();
 193         final ShardDataTreeCohort cohort = dataTree.newReadyCohort(txId, message.getModification(),
 194                 message.getParticipatingShardNames());
 195         final CohortEntry cohortEntry = CohortEntry.createReady(cohort, DataStoreVersions.CURRENT_VERSION);
 196         cohortCache.put(cohortEntry.getTransactionId(), cohortEntry);
 197         cohortEntry.setDoImmediateCommit(message.isDoCommitOnReady());
 198
 199         log.debug("{}: Applying local modifications for Tx {}", name, txId);
 200
 201         if (message.isDoCommitOnReady()) {
 202             cohortEntry.setReplySender(sender);
 203             cohortEntry.setShard(shard);
 204             handleCanCommit(cohortEntry);
 205         } else {
 206             sender.tell(readyTransactionReply(shard.self()), shard.self());
 207         }
 208     }
 209
 210     Collection<BatchedModifications> createForwardedBatchedModifications(final BatchedModifications from,
 211             final int maxModificationsPerBatch) {
 212         CohortEntry cohortEntry = cohortCache.remove(from.getTransactionId());
 213         if (cohortEntry == null || cohortEntry.getTransaction() == null) {
 214             return Collections.singletonList(from);
 215         }
 216
 217         cohortEntry.applyModifications(from.getModifications());
 218
 219         final LinkedList<BatchedModifications> newModifications = new LinkedList<>();
 220         cohortEntry.getTransaction().getSnapshot().applyToCursor(new AbstractBatchedModificationsCursor() {
 221             @Override
 222             protected BatchedModifications getModifications() {
 223                 if (newModifications.isEmpty()
 224                         || newModifications.getLast().getModifications().size() >= maxModificationsPerBatch) {
 225                     newModifications.add(new BatchedModifications(from.getTransactionId(), from.getVersion()));
 226                 }
 227
 228                 return newModifications.getLast();
 229             }
 230         });
 231
 232         BatchedModifications last = newModifications.getLast();
 233         last.setDoCommitOnReady(from.isDoCommitOnReady());
 234         if (from.isReady()) {
 235             last.setReady(from.getParticipatingShardNames());
 236         }
 237         last.setTotalMessagesSent(newModifications.size());
 238         return newModifications;
 239     }
 240
 241     private void handleCanCommit(final CohortEntry cohortEntry) {
 242         cohortEntry.canCommit(new FutureCallback<>() {
 243             @Override
 244             public void onSuccess(final Empty result) {
 245                 log.debug("{}: canCommit for {}: success", name, cohortEntry.getTransactionId());
 246
 247                 if (cohortEntry.isDoImmediateCommit()) {
 248                     doCommit(cohortEntry);
 249                 } else {
 250                     cohortEntry.getReplySender().tell(
 251                         CanCommitTransactionReply.yes(cohortEntry.getClientVersion()).toSerializable(),
 252                         cohortEntry.getShard().self());
 253                 }
 254             }
 255
 256             @Override
 257             public void onFailure(final Throwable failure) {
 258                 log.debug("{}: An exception occurred during canCommit for {}", name, cohortEntry.getTransactionId(),
 259                     failure);
 260
 261                 cohortCache.remove(cohortEntry.getTransactionId());
 262                 cohortEntry.getReplySender().tell(new Failure(failure), cohortEntry.getShard().self());
 263             }
 264         });
 265     }
 266
 267     /**
 268      * This method handles the canCommit phase for a transaction.
 269      *
 270      * @param transactionID the ID of the transaction to canCommit
 271      * @param sender the actor to which to send the response
 272      * @param shard the transaction's shard actor
 273      */
 274     void handleCanCommit(final Identifier transactionID, final ActorRef sender, final Shard shard) {
 275         // Lookup the cohort entry that was cached previously (or should have been) by
 276         // transactionReady (via the ForwardedReadyTransaction message).
 277         final CohortEntry cohortEntry = cohortCache.get(transactionID);
 278         if (cohortEntry == null) {
 279             // Either canCommit was invoked before ready (shouldn't happen) or a long time passed
 280             // between canCommit and ready and the entry was expired from the cache or it was aborted.
 281             IllegalStateException ex = new IllegalStateException(
 282                     String.format("%s: Cannot canCommit transaction %s - no cohort entry found", name, transactionID));
 283             log.error("{}: Inconsistency during transaction {} canCommit", name, transactionID, ex);
 284             sender.tell(new Failure(ex), shard.self());
 285             return;
 286         }
 287
 288         cohortEntry.setReplySender(sender);
 289         cohortEntry.setShard(shard);
 290
 291         handleCanCommit(cohortEntry);
 292     }
 293
 294     void doCommit(final CohortEntry cohortEntry) {
 295         log.debug("{}: Committing transaction {}", name, cohortEntry.getTransactionId());
 296
 297         // We perform the preCommit phase here atomically with the commit phase. This is an
 298         // optimization to eliminate the overhead of an extra preCommit message. We lose front-end
 299         // coordination of preCommit across shards in case of failure but preCommit should not
 300         // normally fail since we ensure only one concurrent 3-phase commit.
 301         cohortEntry.preCommit(new FutureCallback<DataTreeCandidate>() {
 302             @Override
 303             public void onSuccess(final DataTreeCandidate candidate) {
 304                 finishCommit(cohortEntry.getReplySender(), cohortEntry);
 305             }
 306
 307             @Override
 308             public void onFailure(final Throwable failure) {
 309                 log.error("{} An exception occurred while preCommitting transaction {}", name,
 310                         cohortEntry.getTransactionId(), failure);
 311
 312                 cohortCache.remove(cohortEntry.getTransactionId());
 313                 cohortEntry.getReplySender().tell(new Failure(failure), cohortEntry.getShard().self());
 314             }
 315         });
 316     }
 317
 318     void finishCommit(final @NonNull ActorRef sender, final @NonNull CohortEntry cohortEntry) {
 319         log.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionId());
 320
 321         cohortEntry.commit(new FutureCallback<UnsignedLong>() {
 322             @Override
 323             public void onSuccess(final UnsignedLong result) {
 324                 final TransactionIdentifier txId = cohortEntry.getTransactionId();
 325                 log.debug("{}: Transaction {} committed as {}, sending response to {}", persistenceId(), txId, result,
 326                     sender);
 327
 328                 cohortCache.remove(cohortEntry.getTransactionId());
 329                 sender.tell(CommitTransactionReply.instance(cohortEntry.getClientVersion()).toSerializable(),
 330                     cohortEntry.getShard().self());
 331             }
 332
 333             @Override
 334             public void onFailure(final Throwable failure) {
 335                 final TransactionIdentifier txId = cohortEntry.getTransactionId();
 336                 log.error("{}, An exception occurred while committing transaction {}", persistenceId(), txId, failure);
 337
 338                 cohortCache.remove(cohortEntry.getTransactionId());
 339                 sender.tell(new Failure(failure), cohortEntry.getShard().self());
 340             }
 341         });
 342     }
 343
 344     /**
 345      * This method handles the preCommit and commit phases for a transaction.
 346      *
 347      * @param transactionID the ID of the transaction to commit
 348      * @param sender the actor to which to send the response
 349      * @param shard the transaction's shard actor
 350      */
 351     void handleCommit(final Identifier transactionID, final ActorRef sender, final Shard shard) {
 352         final CohortEntry cohortEntry = cohortCache.get(transactionID);
 353         if (cohortEntry == null) {
 354             // Either a long time passed between canCommit and commit and the entry was expired from the cache
 355             // or it was aborted.
 356             IllegalStateException ex = new IllegalStateException(
 357                     String.format("%s: Cannot commit transaction %s - no cohort entry found", name, transactionID));
 358             log.error("{}: Inconsistency during transaction {} commit", name, transactionID, ex);
 359             sender.tell(new Failure(ex), shard.self());
 360             return;
 361         }
 362
 363         cohortEntry.setReplySender(sender);
 364         doCommit(cohortEntry);
 365     }
 366
 367     @SuppressWarnings("checkstyle:IllegalCatch")
 368     void handleAbort(final Identifier transactionID, final ActorRef sender, final Shard shard) {
 369         CohortEntry cohortEntry = cohortCache.remove(transactionID);
 370         if (cohortEntry == null) {
 371             return;
 372         }
 373
 374         log.debug("{}: Aborting transaction {}", name, transactionID);
 375
 376         final ActorRef self = shard.getSelf();
 377         cohortEntry.abort(new FutureCallback<>() {
 378             @Override
 379             public void onSuccess(final Empty result) {
 380                 if (sender != null) {
 381                     sender.tell(AbortTransactionReply.instance(cohortEntry.getClientVersion()).toSerializable(), self);
 382                 }
 383             }
 384
 385             @Override
 386             public void onFailure(final Throwable failure) {
 387                 log.error("{}: An exception happened during abort", name, failure);
 388
 389                 if (sender != null) {
 390                     sender.tell(new Failure(failure), self);
 391                 }
 392             }
 393         });
 394
 395         shard.getShardMBean().incrementAbortTransactionsCount();
 396     }
 397
 398     void checkForExpiredTransactions(final long timeout, final Shard shard) {
 399         cohortCache.values().removeIf(CohortEntry::isFailed);
 400     }
 401
 402     void abortPendingTransactions(final String reason, final Shard shard) {
 403         final Failure failure = new Failure(new RuntimeException(reason));
 404         Collection<ShardDataTreeCohort> pending = dataTree.getAndClearPendingTransactions();
 405
 406         log.debug("{}: Aborting {} pending queued transactions", name, pending.size());
 407
 408         for (ShardDataTreeCohort cohort : pending) {
 409             CohortEntry cohortEntry = cohortCache.remove(cohort.getIdentifier());
 410             if (cohortEntry == null) {
 411                 continue;
 412             }
 413
 414             if (cohortEntry.getReplySender() != null) {
 415                 cohortEntry.getReplySender().tell(failure, shard.self());
 416             }
 417         }
 418
 419         cohortCache.clear();
 420     }
 421
 422     Collection<?> convertPendingTransactionsToMessages(final int maxModificationsPerBatch) {
 423         final Collection<VersionedExternalizableMessage> messages = new ArrayList<>();
 424         for (ShardDataTreeCohort cohort : dataTree.getAndClearPendingTransactions()) {
 425             CohortEntry cohortEntry = cohortCache.remove(cohort.getIdentifier());
 426             if (cohortEntry == null) {
 427                 continue;
 428             }
 429
 430             final Deque<BatchedModifications> newMessages = new ArrayDeque<>();
 431             cohortEntry.getDataTreeModification().applyToCursor(new AbstractBatchedModificationsCursor() {
 432                 @Override
 433                 protected BatchedModifications getModifications() {
 434                     final BatchedModifications lastBatch = newMessages.peekLast();
 435
 436                     if (lastBatch != null && lastBatch.getModifications().size() >= maxModificationsPerBatch) {
 437                         return lastBatch;
 438                     }
 439
 440                     // Allocate a new message
 441                     final BatchedModifications ret = new BatchedModifications(cohortEntry.getTransactionId(),
 442                         cohortEntry.getClientVersion());
 443                     newMessages.add(ret);
 444                     return ret;
 445                 }
 446             });
 447
 448             final BatchedModifications last = newMessages.peekLast();
 449             if (last != null) {
 450                 final boolean immediate = cohortEntry.isDoImmediateCommit();
 451                 last.setDoCommitOnReady(immediate);
 452                 last.setReady(cohortEntry.getParticipatingShardNames());
 453                 last.setTotalMessagesSent(newMessages.size());
 454
 455                 messages.addAll(newMessages);
 456
 457                 if (!immediate) {
 458                     switch (cohort.getState()) {
 459                         case CAN_COMMIT_COMPLETE:
 460                         case CAN_COMMIT_PENDING:
 461                             messages.add(new CanCommitTransaction(cohortEntry.getTransactionId(),
 462                                 cohortEntry.getClientVersion()));
 463                             break;
 464                         case PRE_COMMIT_COMPLETE:
 465                         case PRE_COMMIT_PENDING:
 466                             messages.add(new CommitTransaction(cohortEntry.getTransactionId(),
 467                                 cohortEntry.getClientVersion()));
 468                             break;
 469                         default:
 470                             break;
 471                     }
 472                 }
 473             }
 474         }
 475
 476         return messages;
 477     }
 478
 479     @VisibleForTesting
 480     void setCohortDecorator(final CohortDecorator cohortDecorator) {
 481         this.cohortDecorator = cohortDecorator;
 482     }
 483 }